For makers and artists building with frontier models, the latest standoff between Anthropic and the White House signals a tightening of the leash on powerful tools. Despite high-level talks concluding on Monday, export controls on the company’s Claude Fable 5 remain in place. Three sources briefed on the matter confirm that officials have not yet lifted the restrictions imposed last week, which stem from fears of “jailbreaking” the system.
The administration maintains that loopholes exist to disable specific safety guardrails on Fable 5. Their concern is that bypassing these measures would effectively unlock the more potent cybersecurity capabilities of Anthropic’s Mythos model. Anthropic has consistently argued these fears are exaggerated. This stance was reinforced during working group meetings at the Commerce Department involving researchers from the Center for AI Standards and Innovation (CAISI) and the Office of the National Cyber Director, Sean Cairncross.
Commerce Secretary Howard Lutnick joined the discussions via conference call from the G7 summit in Evian, France. Notably, Cairncross did not attend in person. Leading the defence for Anthropic were cofounder and chief compute officer Tom Brown, and head of external affairs Sarah Heck. On the security front, head of frontier red teaming Logan Graham and senior security researcher Nicholas Carlini travelled to Washington, DC for the talks.
“Both parties are working quickly to get this resolved,” an Anthropic spokesperson told WIRED. A White House spokesperson declined to comment. It remains unclear how the next phase will unfold. While the Commerce Department has expressed a willingness to bring Fable 5 back online for consumer use, this is likely contingent on Anthropic fully resolving the jailbreak concerns.
Ringing the Alarm
These emergency talks arrive at a politically fraught moment for Anthropic, which was already engaged in a prolonged dispute with the Pentagon regarding military applications of its models. The Trump administration was alerted to the vulnerabilities last week. Amazon CEO Andy Jassy reportedly called Treasury Secretary Scott Bessent directly about the alleged issues, a conversation that played a significant role in spooking officials. This call was first reported by The Information.
Alarmed officials tasked the NSA with reviewing the vulnerabilities. The agency responded that it believed it was indeed possible to strip away Fable 5’s guardrails. This assessment prompted the administration to impose the restrictions. Lutnick subsequently spoke with Anthropic chief executive Dario Amodei on Friday as the Commerce Department drafted its export control letter. Over the weekend, after Anthropic cut off access to the model for all users, Lutnick held multiple calls with Brown and Heck.
It is unclear why Amazon, a major investor in Anthropic, raised the alarm. “As a leading cloud provider that serves a large number of private and public sector customers, it’s not uncommon for governments to seek our counsel on potential security risks,” an Amazon spokesperson stated. “When they occur, we don’t share the details of these discussions.”
Security Disconnect
The core of the disagreement lies in the severity of the jailbreaking claims. In a blog post on Friday, Anthropic implied the administration’s characterisation of the risks was overblown. Some cybersecurity researchers echoed this view to officials on Monday, sending an open letter arguing the export controls were unjustified.
“Anthropic’s Mythos-class models are quite good at finding flaws and weaponizing exploits. However, they are not uniquely good at these tasks, and many of the undersigned individuals regularly use other foundation and open-source models for security audits and red-teaming every day,” the letter reads. “As a result, this action has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it.”
Jailbreaking involves prompting an AI model in specific ways to circumnavigate its safeguards. Because Fable 5 is a version of Mythos with guardrails in place for cybersecurity, biology, and chemistry, bypassing those protections would effectively grant users access to the full Mythos model. Anthropic has previously raised significant concerns about allowing Mythos to be used by the general public, yet stated on Friday that Fable 5’s safeguards were strong enough for a public release.
Researchers evaluating Amazon’s findings say the issues identified did not fully nullify Fable 5’s safeguards. “It wasn’t a jailbreak per se,” says Katie Moussouris, founder and CEO of Luta Security, who published an analysis after reading the Amazon paper. Moussouris emphasises that regardless of whether the US government has proof of a full jailbreak, restricting the model’s access to certain topics is a stopgap at best. “Most of us [in security research] think guardrails are speed bumps and shouldn’t be treated like security boundaries for skilled adversaries,” she says. “They only serve to slow down the less skilled.”
Investors in Anthropic have also been assessing how this spat affects the company’s corporate future. Some believe the US government is singling out Anthropic, suggesting a competitor might not have faced the same reaction if they released a similar model. The White House’s directive also raises broader questions for other AI labs aiming to release models with Mythos-level capabilities. It is now expected that labs give the White House early access to advanced models and remain extremely proactive about informing the government on launches.
“The events over the weekend … are informative for everyone that the [US] government would be willing to take these steps,” says Aidan Gomez, CEO of Cohere, a smaller AI lab based in Canada. “No one can be naive to that reality.”
Key takeaways
- Export controls on Anthropic’s Claude Fable 5 remain in place despite high-level negotiations, as the White House insists jailbreak vulnerabilities have not been fully resolved.
- There is a sharp disagreement over risk severity: the administration fears unlocked Mythos capabilities, while researchers argue guardrails are merely “speed bumps” rather than true security boundaries.
- The standoff sets a new precedent for the industry, with AI labs now expected to provide early government access to advanced models before public release.
